charles-accent-heroes
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The Python script (
scripts/heroes.py) utilizessubprocess.runto execute a local Swift script (scripts/ocr.swift) for performing optical character recognition on generated images. This execution is scoped to local files and is a legitimate part of the skill's quality assurance process. - [EXTERNAL_DOWNLOADS]: The skill makes network requests to OpenAI's official API (
api.openai.com) to generate and edit images. These operations are performed using therequestslibrary and target a well-known, trusted service required for the skill's core functionality. - [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided reference screenshots and configuration data. While this represents a potential surface for indirect prompt injection, the risk is mitigated by explicit instructions to the AI agent to use these images only for layout and composition, replacing all branding and text with fictional content. The OCR verification step provides a secondary check to ensure the output conforms to intended instructions.
- [OBFUSCATION]: A translation table for Cyrillic and Greek homoglyphs is present in the Python script. This is used defensively to normalize text recognized by the OCR engine (which frequently misidentifies look-alike characters) and is not an attempt at malicious obfuscation.
Audit Metadata